Research On UAV Sensing Image Mosaic Algorithm

For its low-cost,low-risk,high-resolution,high-efficiency,high flexibility and other advantages,UAV remote sensing system get a widely range of applications in urban planning and municipal administration,the military battlefield reconnaissance,natural disasters,regional assessments and other fields.Yet at the same time,it also has the features of image captured drone,single image space limited,high overlapped,larger data,irregular form.However UAV image stitching technology becomes the bottleneck that restricts its effective application,so it is urgent to make research on a high-quality fast panorama stitching methods.Due to practical needs,image stitching has been a hot research topic,and gradually forms a number of more mature theoretical methods and techniques.Through the study of these theories and technology,this article grasps the concepts and basic image stitching process,summarizes the comparative analysis of different methods of image stitching,considering the characteristics of the UAV images,focuses on the mosaic method based on feature points.Mosaic method based on feature points,has a few steps,including image preprocessing,feature extraction,feature matching,transformation matrix and image fusion.The algorithm of each step of image stitching,contributes to the overall accuracy and speed of image stitching.Image preprocessing including image geometric correction and image noise reduction,image preprocessing,can effectively improve the image registration accuracy.Suitable feature extraction method can improve the extraction of feature points,and can accurately and effectively compress the number of feature points,reducing the amount of computation and improve the speed of image stitching.For the extracted feature points matching we should select the appropriate search strategy in order to make matching efficient and accurate.Matrix coordinate transformation model is to achieve unity between the images to be spliced,our goal is to find as many points as possible to adapt the model,a good transformation model can guarantee the accuracy of registration.The image fusion is to eliminate the different brightness between the different images,in case after the transition zone there are gaps in mosaic image,to enable a smooth and natural transition between images.In this paper,methods commonly used preprocessing method,feature extraction method,feature matching method,matrix transformation model classification and several image fusion methods are introduced and explained.This article focuses on three currently widely used extraction methods,Harris corner detection,SIFT feature extraction operator,SURF feature extraction operator,and conducts research in-depth and make analysis of these feature extraction methods,using VC ++ language,VC6.0 development tools to write programs for image feature extraction experiments,by KD tree search strategy to search for matching pairs,using RANSAC method to make matching purification and build a matrix transformation model,and using direct filling method and fade weighted average algorithm,separately do contrast seamless fusion experiments.And discusses the impact of different reference FIG options and different splicing order on the mosaic effect in the splicing process.Finally,basing on the experiments above,on the MFC platform,developed the UAV image stitching system with interface.The system has the superiorities of faster stitching speed,high stitching image quality,intuitive display.